A Guide On How To Forecast An Election
On the morning of first debate day, “Election Twitter” had a full blown meltdown when Nate Silver’s infamous Five Thirty Eight forecast claimed the election was too close to call. This prompted fellow forecasters to debate him on social media on his models’ findings, arguing that Clinton was still ahead in a competitive election even as Trump surged in the polls. This lead to some taking sides, questioning which forecast was correct. Of course, the choice of sides were made depending who they wished was actually ahead, but even some Clinton supporters or #NeverTrump types wondered if a guy who got all fifty states right in 2012 could be wrong this time.
But the dirty secret is that Silver, while someone with a pretty good record, is not the only good forecaster out there; and his models are not the only ones that exist. In fact there are various ways one could try and forecast an election. Some even use zero polls to do so (and have good records!). The Buckley Club’s “Forecast of Forecasters” is currently tracking over 60 different ones (Article No Longer Exists). So as everyone frets what will happen on election day in the final weeks of the campaign, here is my guide to the various forecasters out there and how to read what they seem to be telling us regarding November 8th…
Forecast Type 1: Polling Based Models
This is the most common type of forecast out there and the most famous of forecasters use these. The concept is simple, use polling to try and predict elections. This could include 50 state surveys that Google, Reuters, The Washington Post, and UPI use. It can also include poll aggregates that try to average polling to get a prediction like Election Projection, Real Clear Politics, or Pollster. Another option in how to use polling to make a forecast is to use the numbers to simulate the potential results and consider the most common result or average results to get a prediction such as 270ToWin, NYT’s Upshot, Sam Wang’s PEC, and of course the already mentioned Five Thirty Eight models from Nate Silver. The strengths of these models is that they react in real time to shifts in the campaign with every news cycle. The weakness is that a temporary surge by the would be losing candidate makes the long term forecast look volatile and weak.
Forecast Type 2: Prediction Markets
Believe it or not, there are people out there that bet money on elections. And they actually have a history of mostly being right and having stable long term outlooks of considering the eventual winner the favorite. Their market choices can be very helpful in predicting elections. While most were still wondering if Trump could win the nomination early on into the first contest, markets were shifting towards considering him the favorite; and even as polls claimed the 2015 UK election would be close enough that Labour could get to government with a coalition, the markets consistently felt the Tories had the edge. The strength of such a method is that the volatility is taken away, the long term forecasts looks great but if conventional wisdom sets in too deeply the markets could be caught off guard by a late surge for the eventual winner if they start out behind. Brexit is a great example of this.
Forecast Type 3 And 4: Econometric And Index Models
But some forecasters use data that have nothing to do with polling or money. Some use what history tells them about biographical information, election cycles, incumbent approvals, economic conditions, and even such things as party primary turnout/vote results to predict an election. These type of models are the least known of all, but every year professors and professional pundits create these and depending on the model some actually have pretty good records. Because these forecasts are made for one or two times in the middle of the campaign, they’re long term stable with not much volatility. However because they are so rarely updated (if at all for some), any surge to the lead by a formerly losing candidate may not be caught. Also, some of these models could be off in an abnormal year like this where they seem to suggest this should be a Republican year, but the candidate quality they ignore can’t account for says otherwise.
Forecast Type 5: Voter Expectations
Unfortunately so much focus is done on the topline results in polling regarding the horserace, that one incredibly predictive question is ignored far too often. Pollsters will ask respondents at times who, regardless who they will vote for, they predict who will win the election. You can see this in the USC/LA Times poll which has gone against other pollsters and predicted a Trump win based on who their respondents claim they’ll vote for, BUT from start to present day those same respondents have said they believe Clinton will win. It’s an interesting question and as mentioned, it has a scary accurate record. But it’s not just polling where you can find this, voters try and predict results through prediction contests set up by other forecasters and historically these “crowd-sourced” forecasts do pretty well too. It only makes sense that voters who are going to actually decide an election be able to predict who will win it.
Forecast Type 6: Expert Thoughts
And finally you have the expert pundit who tries to predict an election based on their personal intuition as election followers. This includes folks like Charlie Cook, Stuart Rothenberg, and Larry Sabato. The weakness of these forecasts though is that these experts tend to cop out and consider close states/seats “Tossups” that go without a prediction until election eve or election morning. However there are panels of experts who try and predict elections with regular updates to their forecast as well.
How To Read What’s Going On
So with so many ways to try and predict an election, how does one best gauge what will likely happen on election night? Well just as it’s foolish to base expectations on just one poll over an average, the same could be said about forecasts. I say this as someone who tries to use his own forecast to try and predict elections, but there WILL be some forecasters that get this election wrong just as there will be polls that botch certain states/races. Therefore looking at what the average forecasts are telling us is the best way to look at where these races are. Which is what forecast aggregates like the one here in The Buckley Club (Article No Longer Exists) are for. As tempting as it is to react to one forecast that may have a daily update claiming the race is too close to call, if a vast majority of the other forecasters are saying not so then it may be wiser to take a step back and consider that forecast is wrong…or may be the only one who gets it right.